Using machine learning techniques for occupancy-prediction-based cooling control in office buildings

Peng, Yuzhen; Rysanek, Adam; Nagy, Zoltán; Schlueter, Arno

doi:10.1016/j.apenergy.2017.12.002

Cited by 250 publications

(90 citation statements)

References 25 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Namely, Chang's model [27] to simulate the occupancy state of a space, Page's model [28] to simulate the number of occupants in a space, and Wang's model [29] to generate the spatial location of each occupant and the space-level occupancy for the whole building. As part of IEA EBC Annex 66 and Annex 79 data-based models and agent-based models for building occupancy have been developed [30,31,32,33,34,35,36]. In [31], data-mining methods were used to derive office occupancy schedules from appliance power consumption measurements.…”

Section: Advanced Occupant Behavior Modelsmentioning

confidence: 99%

“…In [32], data-mining methods were used to derive archetypal working profiles of individual occupants from measured occupancy data of 16 private offices with single or dual occupancy. In [35], machine learning techniques were used for daily occupancy patterns recognition for improving the energy efficiency of an office building. An agent-based model for office buildings has been developed by [36].…”

Section: Advanced Occupant Behavior Modelsmentioning

confidence: 99%

See 1 more Smart Citation

Context-specific urban occupancy modeling using location-based services data

Happle

Fonseca

Schlueter

2020

Building and Environment

View full text Add to dashboard Cite

Energy-related occupant behavior is a major source of uncertainty in building and urban energy performance simulations. Standardized assumptions, published by ASHRAE and others in the form of occupancy schedules, are widely used in research and practice, especially on the district-scale. In this work, we gathered location-based services data to create context-specific, data-driven occupancy schedules. Using a web mapping service, we collected data for retail and restaurant uses in the downtown neighborhoods of 13 different U.S. cities to create data-driven schedules for each context. The schedules were compared to ASHRAE standard assumptions using the earth mover's distance approach and the schedules' energy-related features. We found that standard schedules seem to significantly overestimate weekly building occupancy, although the shapes of the schedules are generally similar. The use of standard schedules could therefore, have significant impacts on district-scale energy demand simulations, as the overestimation will be cumulative.As compared to the differences between data-driven and standard schedules, the differences between different locations are significantly smaller. However in extreme cases, the weekly cumulative occupancy and the number of occupied hours differ by more than 30% between locations, which means that context-specific differences together with climatic differences might also impact building performance simulation results. Furthermore, we found differences in daily data between the different days of the week. In particular, the observed behavior on Fridays is significantly different from other weekdays for both considered use-types. This indicates that the conventional categorization of occupant behavior models into three day-types: weekday, Saturday, and Sunday, should be reconsidered.

show abstract

Section: Advanced Occupant Behavior Modelsmentioning

confidence: 99%

Section: Advanced Occupant Behavior Modelsmentioning

confidence: 99%

Context-specific urban occupancy modeling using location-based services data

Happle

Fonseca

Schlueter

2020

Building and Environment

View full text Add to dashboard Cite

show abstract

“…The ML based building load forecast models have been extensively reviewed in [8][9][10][11][12][13]. Some of these models are developed for specific application areas such as building performance measurement and verification [14][15][16][17], building control [18][19][20] and demand-side management [21,22], whereas a significant number of studies are application agnostic. Literature demonstrates the capability of supervised ML algorithms such as artificial neural networks (ANN) [23], support vector machines (SVM) [24], decision trees [25,26], Gaussian processes [27][28][29] and nearest neighbours [30], among others, in developing reliable building load forecast models.…”

Section: Literature Reviewmentioning

confidence: 99%

A Machine Learning Pipeline for Demand Response Capacity Scheduling

Krishnadas¹,

Kiprakis

2020

Energies

View full text Add to dashboard Cite

Demand response (DR) is an integral component of smart grid operations that offers the necessary flexibility to support its decarbonisation. In incentive-based DR programs, deviations from the scheduled DR capacity affect the grid’s energy balance and result in revenue losses for the DR participants. This issue aggravates with increasing DR delivery from participants such as large consumer buildings who have limited standard methods to follow for DR capacity scheduling. Load curtailment based DR capacity availability from such consumers can be forecasted reliably with the help of supervised machine learning (ML) models. This study demonstrates the development of data-driven ML based total and flexible load forecast models for a retail building. The ML model development tasks such as data pre-processing, training-testing dataset preparation, cross-validation, algorithm selection, hyperparameter optimisation, feature ranking, model selection and model evaluation are guided by deployment-centric design criteria such as reliability, computational efficiency and scalability. Based on the selected performance metrics, the day-ahead and week-ahead ML based load forecast models developed for the retail building are shown to outperform the timeseries persistence models used for benchmarking. Furthermore, the deployment of these models for DR capacity scheduling is proposed as an ML pipeline that can be realised with the help of ML workflows, computational resources as well as systems for monitoring and visualisation. The ML pipeline ensures faster, cost-effective and large-scale deployment of forecast models that support reliable DR capacity scheduling without affecting the grid’s energy balance. Minimisation of revenue losses encourages increased DR participation from large consumer buildings, ensuring further flexibility in the smart grid.

show abstract

“…As a conclusion from the presented studies, the human factor is identified as a cause of the gap between the predicted and measured energy consumption in buildings. Resultantly, there was a number of research studies that proposed predictive models of OB in terms of the occupant's manual control of sun shades ( [10], [11]), use of air conditioning ( [12], [13], [14], [15], [12]), plugin loads ( [16], [17], [18]) and window openings.…”

Section: Obmentioning

confidence: 99%

Window opening model using deep learning methods

Markovic

Grintal

Wölki

et al. 2018

Building and Environment

View full text Add to dashboard Cite

Occupant behavior (OB) and in particular window openings need to be considered in building performance simulation (BPS), in order to realistically model the indoor climate and energy consumption for heating ventilation and air conditioning (HVAC). However, the proposed OB window opening models are often biased towards the over-represented class where windows remained closed. In addition, they require tuning for each occupant which can not be efficiently scaled to the increased number of occupants. This paper presents a window opening model for commercial buildings using deep learning methods. The model is trained using data from occupants from an office building in Germany.In total, the model is evaluated using almost 20 mio. data points from 3 independent buildings, located in Aachen, Frankfurt and Philadelphia. Eventually, the results of 3100 core hours of model development are summarized, which makes this study the largest of its kind in window states modeling. Additionally, the practical potential of the proposed model was tested by incorporating it in the Modelica-based thermal building simulation. The resulting evaluation accuracy and F1 scores on the office buildings ranged between 86-89 % and 0.53-0.65 respectively. The performance dropped around 15 % points in case of sparse input data, while the F1 score remained high.

show abstract

Using machine learning techniques for occupancy-prediction-based cooling control in office buildings

Cited by 250 publications

References 25 publications

Context-specific urban occupancy modeling using location-based services data

Context-specific urban occupancy modeling using location-based services data

A Machine Learning Pipeline for Demand Response Capacity Scheduling

Window opening model using deep learning methods

Contact Info

Product

Resources

About